A new total generalized variation induced spatial difference prior model for variational pansharpening

ABSTRACT

This letter proposed a new and effective total generalized variation (TGV) induced spatial difference prior model for variational pansharpening problem, which aimed to estimate a high-resolution (HR) multispectral (MS) image from its low-resolution (LR) version and the corresponding HR panchromatic (Pan) image of the same earth scene. In addition to using the local spectral consistency constraint for spectral information preserving, this letter particularly exploited the spatial difference prior between the HR-MS and Pan images, and hence proposed a new TGV-induced spatial difference prior term for spatial information preserving. Then, an efficient optimization algorithm was designed for solving the proposed model under the fast iterative shrinkage-thresholding algorithm (FISTA) framework. Finally, the experimental results validated that the proposed method performed higher spatial and spectral qualities than various methods in both the subjective and objective aspects.

Additional information

Funding

This work was supported by the National Natural Science Foundation of China under Grant 61802202, by the Natural Science Foundation of Jiangsu Province under Grant BK20170905, and by the NUPTSF under Grant NY218025 and Grant NY217137.